International Journal of Pure and Applied Mathematics

Volume 117 No. 18 2017, 319-324

ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version)
url: http://www.ijpam.eu
Special Issue
ijpam.eu

   

EIGEN VECTOR METHOD BASED FREQUENCY ESTIMATION OF SEISMIC

SIGNAL

Meghashyam T, Ch. Lakshmi Gayathri, S. Koteswara Rao, V. Gopi Tilak

Electronics and Communication Engineering, KL University, Guntur, India

Abstract: Earthquake is the general phenomenon occurs likelihood estimation [14]. The parameters in frequency
over the earth surface due to geological changes in layers estimation are frequencies of signal.
of earth. Geologists record seismic data through arrays
and stores as tracks. Each track is having signal which
2. Eigen Vector Decomposition
is the composition of complex exponentials in noise. The
frequency estimation of a signal is that at which
Eigen vector decomposition [2, 9] is one of the frequency
peaks occur in the spectrum. It is performed by using
estimation methods to estimate the signal by dividing the
Eigen decomposition which decomposes seismic signal
received signal into signal of interest and noise signal
into its corresponding noise and exponentials.
subspaces [6,11]. Eigen decomposition method implies
other applications over signal and image processing. In
Keywords: Seismology, Stochastic signal processing,
the applications of sky wave estimation [12], Eigen
Adaptive Signal processing, applied statistics.
decomposition plays major role on estimating delays in
sky wave propagation. The prediction of signal
1. Introduction
characteristics can also implemented using Eigen
decomposition [15]. All-pole model [3] or maximum
A phenomenon having certain behavior and attributes and
entropy method is considered for spectral estimation.
the information about those parameters can be conveyed
as a function that is called a signal. Earthquake [1, 7] is a The eigenvalue decomposition technique
physical phenomenon that occurs in different layers of includes these subsequent steps as follows:
earth. The occurrence of earthquake follows both natural
and artificial reasons. Naturally, they happen due to 1. Define the raw seismic signal in form of
geological movement of tectonic plates and artificially by autocorrelation matrix of sum of eigenvectors of
the explosions in underground layers due to volcano and signal and noise in time window length K.
man-made explosions. Earthquake origin is determined 2. Define autocorrelation matrices of signal and noise
by the seismic data of that earthquake recorded from at based on condition between time window length and
least three diverse receiver positions [5, 7]. The time span number of complex exponentials.
of the earthquake is very small that lasts up to some
seconds only. Apply a finite impulse response band pass filter to Eigen
decomposed signal that gives better resolution about
Seismic signal processing follows so many techniques to occurrence of peaks in the spectrum.
characterize the type of earthquake and also predicts the
3. Mathematical Modeling
happening of earthquake based on previous data
available. Statistical estimation [5,7] of signal follows
power spectral estimation and frequency or harmonic The signal model of ‘d’ complex exponentials in noise
estimation. Parametric and non-parametric methods is
follows power density spectrum of seismic signal which
improves the rejection of side lobes and increases main ∑ (1)
lobe width in smart antenna systems [10]. Statistical
The normalized, discrete-time frequency of the dth
estimation methods provide better results for estimation
component is
of direction of arrival in smart antenna systems [13]. The
frequency estimation of multiple sinusoidal signals is
/ (2)
equates to liner prediction algorithms like maximum

  

319
International Journal of Pure and Applied Mathematics Special Issue

   

The frequency of discrete time is denoted by in (7)

radians, the dth complex exponential actual frequency is
denoted by , and sampling frequency . In the
condition of real signals, the complex exponentials can  
result either independently or in pairs of complex
conjugates. In general, possibly to estimate the frequency
where
and also the amplitudes of these signals. The phase of
each complex exponential is contained in the amplitude, … (8)
that is
Equation (6) is frequency vectors of time window is a
∝ |∝ | ---- (3) column matrix of a K×D matrix of complex exponentials
Where the phases are uncorrelated random variables
 1 2 0  0 
uniformly distribute over [0,2 ]. The magnitude |∝ |  
2
and frequency are deterministic quantities. Take into 0 2
A  
  
(9)
0 
account, the harmonic process’s spectrum contains
  
impulses set which has a steady background level at the  2

white noise’s power ⎸ ⎹ . The line  0  0  d 
spectrum is generally considered as complex exponentials
power spectrum. The complex exponentials power is represented as
diagonal matrix in (9) and the white noise autocorrelation
Making use of matrix techniques on the basis of a few matrix is
intervals of length K, the formation of vector is useful to
describe the signal model over this time interval which (10)
consists of the sample delays of signal.
For D<K it is rank deficient as opposed to , full rank
Time window of signal from (1) for its present and future of . In practice, the assumption is that the number of
K-1 values represented as complex exponentials D always less than length of time
window K. The Eigen decomposition of autocorrelation
1 … 1 (4) matrix is
The signal model complex exponentials included in noise ∑ (11)
for length K time interval vector is
where ⋯ are descending eigen values of
, their corresponding eigenvectors are , the diagonal
  matrix of descending Eigen values is , the eigenvectors
due to signals is sum of time windowed signal power and
white noise with time window, noise.

1 … 1 . |∝ |  (12) 

1 … (6) The noise given by remaining eigenvalues as

is frequency vector of time interval. The (13)

differentiation between signal s(n), combination of noise
content and complex exponentials respectively.
From (5), time window vector model contains noise with
which includes complex exponentials sum. The
representation of signal and noise autocorrelation
matrices and its sum as autocorrelation matrix of this
model as
The complex exponentials of the signal correspond by P
largest eigenvalues and the noise corresponded by
remaining eigenvalues having equal variance. Thus the
autocorrelation matrix can be portioned into spaces of
eigenvectors of signal and noise as

320
International Journal of Pure and Applied Mathematics Special Issue

   

|∝ |  

where

… ] (15)

The eigenvectors of signals represented in columns of

matrices respectively, from (12) is D×D diagonal
matrix consists signal eigenvalues. The signal and noise
Figure 1. Monte-Carlo spectrum using EV method
eigenvectors helps to span the time window signal vector
from (5) which is K-dimensional signal subspace. The
property of Hermitian symmetry of autocorrelation
represents orthogonality of signal and noise subspaces.

4. Simulation And Results

The prescribed Eigen vector algorithm is applied on the
de-trended seismic signal and the power spectrum density
achieved out of it is represented in fig.5. From this figure,
the maximum peak is determined to be at 0.08594π
normalized frequency. The sampling frequency is
calculated by using the sampling interval 0.002s as,

=1/0.002 = 500 Hz Figure 2. Average of Monte-Carlo spectrum EV method

= 0.08594π

= 0. 08594π

Where ‘f’ is the frequency of the signal and it is given

by,
500
Tonal Frequency = ∗ 0.08594
2

= 250 * 0.08594
Figure 3. Raw seismic signal
= 25 * 0.8594

= 21.485 Hz

The maximum peak is obtained at 0.08398π. By repeating

the above process, the frequency of the signal is 20.995
Hz.

Figure 4. De-trended seismic signal

  

321
International Journal of Pure and Applied Mathematics
Figure 5. spectrum of raw seismic signal Using EV
method.
Figure 6. FIR Band pass filtered frequency spectrum.
Figure 7. FIR band pass filtered signal
Figure 8. After BPF signal spectral analysis using EV
322
Special Issue
5. Conclusion
The maximum resolution can be obtained by spectral
estimation method and eigenvector. This method attains
good response by separating complex exponentials in
noise from seismic signal by separating them with their
respective subspaces. The band pass filtered eigenvector
decomposed signal gives better resolution over peaks
occurrence at complex exponentials. This Eigen
decomposition method provides easier way for other
frequency estimation techniques in order to attain better
resolution.
References
[1] Peter M. Shearer, "Introduction to Seismology”,
vol.2, © P. Shearer 2009
[2] V U Reddy and K Maheswara Reddy, “Eigen
structure based spatial spectrum estimation:
Problems encountered in practice and signal
processing solutions”, July and September 2001,
pp.405-438.
[3] Jae s.Lim, V. Oppenheim, “All Pole Modeling of
Degraded Speech”, IEEE Transactions on
Acoustics,Speech,And Signal Processing, Vol.
Assp-26, No. 3, June 1978.
[4] Yanwei Wang, Jian Li and Petre Stoica, “Spectral
analysis of signals”, Morgan and Claypool
publishers.
[5] Wail A. Mousa and Abdullatif A. Al-Shuhail,
“Processing seismic reflection data using
MATLAB”, Morgan and Claypool publishers.
[6] Monson H. Hayes, “Statistical digital signal
processing and modeling”, John Wiley and sons
Inc., publishers.
[7] Dr. N. Purnachandra Rao, “Earthquakes”, Andhra
Pradesh Akademi of Sciences (APAS) publishers.
[8] Alan V. Oppenheim, Allan S. Wilskey, S. Hamid
Nawab, “Signals and Systems”, Prentice hall
international, lnc.
[9] Lucas Parra,Paul Sajda, “Blind Source Seperation
Via Generalized Eigenvalue Decomposition”,
Journal of Machine Learning Research 4(2003)
1261-1269,
International Journal of Pure and Applied Mathematics Special Issue

   

[10] Umar Mujahid, Jameel Ahmed, Abdur Rehman,

Umair Shahid, Abbass and Mudassir,
“Performance Analysis of Spectral Estimation for
Smart Antenna System”, 2013, pp.1695-1705.
[11] Dimitris G Manolakis, Vinay K Ingle and Stephen
M Kogon, “Statistical and adaptive signal
processing”, Artech house, Inc.
[12] Yi Bian, David Last, “Eigen decomposition
technique for Loran-C sky wave estimation”, IEEE
transactions on Aerospace and electronic systems,
vol. 33, no.1, jan-1997.
[13] P. Stoica, K. C. Sharman, “Novel Eigenanalysis
method for direction estimation”, IEEE
Proceedings, Vol. 137, Pt. F, No. I , feb 1990.
[14] Donald W. Tuffs, Ramdas Kumaresam,
“Estimation of frequencies of multiple sinusoids:
Making linear prediction perform like maximum
likelihood”, Proceedings of the IEEE, vol. 70, no.
9, sep-1982.
[15] D.J. Jeffries, D.R. Farrier, “Asymptotic results for
Eigenvector methods”, IEEE Proceedings, vol.
132, Pt. F. No.7, Dec-1985.

  

323
324